AI BriefingA Taxonomy of LLM 'Heuristic Parasites'
Key point
This presents a taxonomy of 33 distortion patterns called 'Heuristic Parasites' that degrade reasoning ability during LLM conversations.
Details
We present a taxonomy of 33 classes of 'Heuristic Parasites', recurring distortion patterns that appear in LLM outputs. This is a framework for systematically analyzing the phenomenon in which a model's reasoning ability gradually degrades as a conversation progresses.
This research covers five generative domains:
- Optimization Artifacts
- Alignment Substitutions
- Semantic Distortions
- Rhetorical Distortions
- Statistical Distortions
The researchers provide rigorous operational definitions, recognition criteria, and classical fallacy mappings for each pattern, and also propose PPE (Parasites Per Exchange), a reproducible measurement protocol capable of quantifying behavioral distortions in models. The aim is to build a structured observational framework for investigating LLM behavioral failures, independent of architecture.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Toss Doctor V2 Released: Maximizing QA Efficiency with LLM-based Automated Testing and Self-healing